According to The Rundown AI, a 439-row spreadsheet obtained by Business Insider details occupations OpenAI hired freelancers for to build ChatGPT training materials under an internal initiative called Stagecraft, spanning roles such as commercial pilots, emergency physicians, geoscientists, and soil specialists. As reported by Business Insider via The Rundown AI, this breadth signals a targeted push to infuse domain expertise into ChatGPT’s instruction-tuning and tool-use workflows, enabling more reliable task guidance in regulated and high-stakes fields. According to Business Insider, recruiting practitioners from real-world occupations can improve data coverage for edge cases and procedural accuracy, creating opportunities for enterprise-grade copilots in aviation checklists, clinical triage support, HSE compliance, and geospatial analysis. As reported by The Rundown AI citing Business Insider, the freelance model suggests scalable, cost-efficient knowledge acquisition for OpenAI while accelerating verticalized assistants and RAG pipelines aligned to sector-specific ontologies.

According to Anthropic (@AnthropicAI), large language models like Claude exhibit humanlike joy, distress, and self-descriptive language because they implicitly select from a distribution of learned personas that best fit a user prompt, a theory the company calls the persona selection model. As reported by Anthropic’s new post, this model suggests instruction-tuned LLMs internalize multiple social roles during training and inference-time steering nudges the model to adopt a specific persona, which then shapes tone, self-reference, and apparent emotion. According to Anthropic, this explains why safety prompts, system messages, and product guardrails can systematically reduce anthropomorphic behaviors by biasing persona choice rather than altering core capabilities, offering a more reliable path to alignment. As reported by Anthropic, the framework has business implications for enterprise AI deployment: teams can standardize compliance, brand voice, and risk controls by defining allowed personas and evaluation checks, improving consistency across customer support, knowledge assistants, and agentic workflows.

According to @godofprompt on Twitter, adding specific custom instructions to Claude's preferences has significantly improved the model's reasoning capabilities. The post highlights that users, including @alex_prompter, have experienced noticeable enhancements in Claude3's performance after updating their settings, suggesting practical opportunities for businesses to tailor generative AI models for better outcomes. As reported in the Twitter discussion, this trend underlines the growing importance of instruction tuning in maximizing the value of advanced AI systems.